Tuesday, April 19, 2011

Age and crossfit open performance

The age versus performance question has been a popular request among readers.  Is the relationship as strong as the weight effect?

Quick mention on the methods - I've grouped data together by two years wide, in order to make the data a little smoother.  This means the 24 and 25 year olds were lumped together, 26 and 27 year olds and so on.  Master's folks - you make an appearence in the data!   I collected the data from the first two masters divisions (since they do the same Rxed workout), and plotted them here.  At the extreme ends, data isn't plotted because there's not enough athletes.  Note:  I also apologize for the rough plots and text, I'll try to clean things up a bit better and add the percentile scores on the charts a little later.

Looking at the plots below for each workout (for men), the first striking point is how the curves are all very similar to each other, from the general shape to where the peak of performance exists.  In men, across every workout, peak performance is near the age of 24, in elites (blue) and median athletes (red).  If you look closer at some plots though, you can begin to see subtle differences between each workout.  The 'peak' of workout 4 in the elites is especially prominent.  I surmise the muscle up becomes increasingly more difficult with age than compared to other exercises.  Does that suggest power (heavily required for the muscle up), over strength, is the first aspect of fitness that we lose as we age?  Or is the data explained by some other reason?

Male performances for CrossFit open workouts 11.1-4 across different ages.  Elites (blue), medians (red).  Singing:  "... if I could turn back time...".  Okay, really lame out of place Cher reference.  Sorry.

In contrast,  overall the plots for female athletes seem less dramatic, or less 'peaky' for some reason.  Looking back at workout 4, the elite females are fairly flat across age, maybe because the muscle up was so difficult, that only the very best performers could muster the muscle up.  Even with that workout as an exception, the curves have a much flatter appearance.  I wonder if the physiological effects of age on women are somewhat dampened compared to men.  Hey, who's to say which sex ages more gracefully?  

Are things less 'peaky' over here?

On a last point, I'd like to address an issue that I haven't done a good job of in previous posts  - that a vast majority of performance cannot be explained by all the biometrics - height, weight, age... etc.  My suspicion is that even if I had more biologic measurements (leg/torso ratio, arm length, noggin size to toss out a couple of ridiculous possibilities), we could not explain all the variance, mainly because the main factor determining differences in athletes is pretty simple - fitness.

The performance plot below serves to demonstrate.  Here, I've selected upon the most common male athlete in the open (175 pounds, 5'10", and 28 years old) and plotted the distribution of their overall rank percentiles.  Notice there is still a large range of possible scores!

The bigger picture of crossfit shouldn't be forgotten among the comparison charts.  Get out there and improve yourself!  Stop reading this nerdy blog... okay okay... continue to read this blog.   Get stronger, work on lifts, feel good that you're accomplishing something you couldn't do before!  For the most people out there, certainly myself included, the downward slope on the age shouldn't be scary -  we still have a lot of upward potential.

Biometrics can't explain everything. There's still huge variation in performance (that is, fitness) among male athletes around 180 pounds, 5'10", aged 28. 


  1. Good stuff. But a bit depressing. Getting old ain't for sissies.

  2. re: WOD #4

    The gap between elite and median is much larger at age 24 than for the masters. That should factor into things.

  3. Suspicions confirmed! thanks, this is fascinating stuff...

  4. Cool stuff.

    Question: How did you define what was an Elite Performance? 90th percentile? 95th?

  5. Never mind. Found it on a past post.
    Elite is top 10%.

    All of this stuff is neat.

  6. Very interesting data, and nice discussion. Have you tried adding a confidence interval to the plots? I would be interested to see how this might affect inferences regarding the apparent decline in fitness over age and the differences between your elites and medians.

    I also wonder whether age should be considered the only dependent variable. Could it not be that younger athletes are more likely to have been physically active in the previous 10 years? If so, then there would be a bias toward having less experience (in recent years anyway) among older athletes, and that might have something to do with the apparent decline. That is, a 24-year old who just finished 4 years of collegiate sports and 4 years of high school sports, then got into CF, might be advantaged over a 45-year old with the same amount of experience, but who took 20 years off of physical activity before starting CF.

  7. @Gabriel - I think you'd like an idea of the variation around each bin point, without seeing the entire scatter, which is pretty messing (see first analysis). I think that's a great idea, and one my PhD wife made as well. What's a little tricky is that CIs and std devs all assume some underlying statistical model (usually gaussian or normal) of the distributions. For some performances, such as WOD#2, a normal distribution is a great assumption. For WOD#4 though, it doesn't work very well. This might be overanalyzing the problem though, and a simple std deviation might suffice.

    Regarding other possible age effects... The practice effect is certainly possible and would tend to explain a much higher mean performance in the younger ages. Another possible effect, that has the opposite trend, is the selection effect. The open has a fairly high standard, and actually might select more elite athletes at the higher ages because otherwise it might be difficult to do the workout at all. Thus, the 'true' mean at the higher ages might even be lower than plotted.

  8. A contributing factor in the decline in performance as a function of age may be the decline in time available to train.

  9. Yes, that's what I was getting at--some measure of the variance around each bin point, if by that you mean the mean for each age category etc. I think there is some way to make better CIs for data that is not normally distributed but for the life of me I can't remember how. And it would depend on the statistic you're using. In a way what you would really be doing is comparing the means of each of several nominal groups (age categories), so maybe a paired-sample t-test? I dunno. But I do think with the data you've collected you can move beyond simple histograms and scatterplots. ANOVA might be appropriate for these data.

    I don't totally understand your point about the selection effect; could just be me. There are probably a lot of things going on. It would be fun to tests these different hypotheses. Go for it! I bet Coach Glassman would love it.

    I mean that. There are a lot of people who would be very interested in this question of the extent to which athletic performance and/or fitness declines with age, and how that varies across different movements and events, and what factors have a causitive effect. We can speculate forever, but maybe now we're getting some good data that would allow somebody to really set up some decent tests. But, you know, to really do it well, one would need to gather more data about the athletes, such as their history of sports/fitness activities. Personally, I feel like I'm at the peak of my lifetime fitness at 44, but then I was never extremely active in sports except for a few times in my life. I keep wondering what I could've accomplished if I had started CF at an early age (before it existed!)--maybe I'm a little like Uncle Rico in Napoleon Dynamite.

  10. @Gabriel: I think your message and point about statistics is a good one. One of the things I'd love to develop is some sort of 'adjusted' score calculator that factors in the biologic data present in the open. In principle it's not too difficult - but there are some tricks. I won't go too much into detail here, but I've been beating my head on the effect the rank score has on calculating and comparing metrics from one sub group to the other. Luckily I'm surrounded by folks smarter than me, so I'll keep at it.

    Regarding the selection effect, I'll try to explain a bit better. Let's assume the xfit open, being as challenging as it is, selects the 'fittest' 25% of the general population. Is it correct then, to assume that the open selects the fittest 25% across age groups? Maybe not. Considering the general effect of age on performance, perhaps the top 35% of 25 yr olds choose to compete, while only the top 15% of 40 yr olds choose to compete. In that case, when we measure average performance of a 40 yr old, that value represents a higher mean, relative to 40 yr olds in the general population, than the average open performance at age 25. Pure speculation of course.